Agglomeration and the adjustmen
t
of the spatial economy
§Pierre Philippe Combes♣ GREQAM
Gilles Duranton♦
University of the Andes and London School of Economics
Henry G. Overman♥ London School of Economics
April 2005
ABSTRACT: We consider the literatures on urban systems and New Economic Geography to examine questions concerning agglomeration and how areas respond to shocks to the economic environment. We first propose a diagrammatic
framework to compare the two approaches. We then use this framework to study a number of extensions and to consider several policy relevant issues.
Key words: Urban systems, New Economic Geography, Urban and regional policy, diagrammatic exposition
JEL classification: R00, R58
§
Financial support from the UK Office of the Deputy Prime Minister is gratefully acknowledged. We are also very grateful to Tony Venables who helped develop our thinking on these issues while working jointly on a related project focused on intercity linkages.
♣
GREQAM, 2, rue de la Charité, 13236 Marseille, France, combes@ehess.univ-mrs.fr. Also affiliated with Paris-jourdan Sciences-Economiques and the Centre for Economic Policy Research.
♦
Facultad de Economía, Universidad de los Andes, Carrera 1 #18A-70, Bogotá, Colombia,
gduranto@uniandes.edu.co. Also affiliated with the Centre for Economic Policy Research and the Centre for Economic Performance at the London School of Economics.
♥
Department of Geography and Environment, London School of Economics, Houghton Street, London WC2A 2AE, United Kingdom, h.g.overman@lse.ac.uk. Also affiliated with the Centre for Economic Policy Research and the Centre for Economic Performance at the London School of Economics.
CEDE
DOCUMENTO CEDE 2005-26
Aglomeración y Ajuste
de la Economía Espacial
Pierre Philippe Combes GREQAM
Gilles Duranton
Universidad de los Andes y London School of Economics
Henry G. Overman London School of Economics
Abril 2005
RESUMEN: Mediante la revisión de la literatura de sistemas urbanos y de la Nueva Geografía Económica, analizamos aspectos relacionados con la aglomeración y cómo las áreas responden a los shocks del entorno económico. Inicialmente proponemos un marco esquemático para comparar las dos propuestas. Luego usamos este marco para analizar extensiones de los modelos y tener en cuenta varios aspectos relevantes para la definición de políticas.
Palabras Claves: Sistemas Urbanos, Nueva Geografía Económica, Política Regional y Urbana, Presentación esquemática
1. Introduction
The last fifteen years have seen a renewal of interest from economists in spatial issues. What started as
an attempt to re-invigorate regional economics by Paul Krugman and his associates has led to a flurry
of theoretical work, which culminated with the publication of Fujita, Krugman and Venables’ Spatial Economy (1999). In turn, these theoretical developments – the ‘New Economic Geography’ – have triggered a wave of empirical research and also inspired more policy-oriented work (e.g., Baldwin,
Forslid, Martin, Ottaviano and Robert-Nicoud, 2003).
Despite this surge in interest, several crucial issues still hinder the development of spatial economics
and its application to a range of policy issues. First, despite some claims to the contrary, the New Economic Geography (NEG) is not the undisputed framework of reference on all things spatial.
Instead, it cohabits somewhat uneasily with the older approach of urban systems theory that builds on
Henderson’s (1974) pioneering work. There have been some attempts to reconcile these two
frameworks (Fujita, Krugman and Mori, 1999, Tabuchi, 1998) but at this point we are still missing a
clean comparison between the two approaches. Second, despite a very large number of extensions, many important dimensions of these models remain under-explored. This problem is particularly acute
from a policy perspective, where the most policy-relevant issues – imperfect labour mobility, local
regulations, labour force participation etc – have attracted the least attention. Third, both urban systems and NEG models are very difficult to work with and are thus still very poorly understood
beyond their narrow circles of contributors, two circles which, somewhat surprisingly, do not exhibit
much overlap. Possibly as a result of these three problems, these approaches have yet to make big
headway in influencing regional and urban policy.
In an attempt to address these issues, this paper proposes a unified diagrammatic framework, which
encompasses both urban systems and NEG. This diagrammatic framework helps highlight the key
similarities and differences between the approaches and is also amenable to extension to give insights
on a range of policy issues. The rest of the paper is structured as follows. In section 2, we first
highlight the key ingredients of any consistent theory of agglomeration and the adjustment of the
spatial economy to shocks. We then introduce our diagrammatic framework. In section 3, we use it to
compare the urban systems approach to NEG and discuss their common ground and highlight their
differences. In section 4, we show that our diagrammatic framework is flexible enough to
accommodate a number of existing extensions to both approaches. In section 5, we apply our
diagrammatic framework to a number of new extensions focusing our attention, in particular, on the
most policy-relevant issues.
replacement of formal models by graphs in this, or any, field of research. Ultimately, any fully-fledged
unified approach to regional and urban economics will have to rely on formal models.1 We only view
our diagrammatic approach as a first step, a step that is particularly useful for the preliminary
exploration of well-articulated problems that can be very difficult to model more formally. In addition,
diagrams are, of course, often very efficient tools with which to explain economic theories to students
and policy-makers.
2. A diagrammatic analytical framework
Most economic theories concerned with agglomeration and how places respond to shocks have a
common underlying structure even if this is not always apparent at first sight. Our objective is first to
identify this common structure and consider the assumptions of any consistent theoretical framework.
Any model of a spatial economy has to contain three elements: A spatial structure, a production
structure, and some assumptions about the mobility of goods and factors.
Spatial structure: Evidently, the geographical space must be divided into some number of units, covering cities or regions in the country (we will call both of these ‘areas’).2 The number of areas is
often taken as given but in practice this need not be the case. New areas can be created either by
private land developers (as in the US) or through local or central government action (as in most of the
EU). Given a selection of areas, the model must define both an external and an internal geography.
The external geography is about how the areas are related to each other, while the internal geography
is concerned with the internal structure of the area (land, housing, infrastructure etc). If the focus is on
linkages between places, the external geography is obviously crucial, as it will determine the channels
through which a change in one area affects others. It is often taken to be exogenous, but this need not
be the case as the distance between areas can change, for example as a result of changes in policy or
technology. A number of factors may also change the internal geography of areas.
Production structure: Assumptions on the local availability of primary and intermediate inputs play a crucial role. Of the primary inputs, most important is labour.3 Detailed modelling of labour is required,
both to capture different skill levels and to allow for labour force participation decisions. Intermediate
inputs may also play a crucial rule. It is possible to specify directly an aggregate production function
1
Among many reasons, we would underscore that checking the consistency of formal models is much easier than that of diagrams. Formal models also lend themselves well to quantitative exercises, which can only be sketched with diagrams.
2
Our focus is only on sub-national applications of these models.
3
relating primary and intermediate inputs to some final output. Often this is not very enlightening.
Instead, assumptions about the nature of products (particularly the degree of product differentiation),
the input-output structure and the degree of firm level increasing returns will determine the extent and
nature of any aggregate increasing returns.4 Imperfect competition, resulting from firm level
indivisibilities, usually looms large in the analysis.
Mobility of goods and factors: Assumptions about mobility, both between and within areas, play a crucial role in determining the spatial structure of the economy. These assumptions should cover the
geographical mobility of goods, services, ideas, technologies, and primary factors. The extent to which
material inputs and outputs are tradable clearly varies across sectors. Some activities are ‘footloose’
while others are necessarily tied closely to the markets that they serve. The mobility of ideas and
technologies will determine how the production function varies across space. Finally, assumptions on
primary factors are crucial. Land is immobile, although its availability for use in different activities
(e.g., housing versus agriculture) is endogenous. Capital is often taken as highly mobile, with the same
supply price in all areas. The (imperfect) mobility of labour is a fundamental issue that warrants
careful treatment. The conjunction of some form of increasing returns with imperfect mobility is the
main driver of the clustering of economic activity and a key determinant of the way in which linkages
between areas operate. Assessing the strength of such clustering, its range across space, and its scope
across industrial sectors is of key importance, as will be clear from our detailed discussion below.
Before continuing, we make three remarks. First, the analysis of linkages between cities and regions is
inherently a ‘general equilibrium’ problem, in which the researcher has to look beyond the direct (or
impact) effect of a change and assess the induced changes that follow. Doing this is possible only if
there is a clear analytical framework within which the various effects interact. Second, the exact level
of spatial disaggregation selected cannot be pre-determined and depends on the questions being
addressed and the availability of data. Third, the usefulness of different levels of sectoral
disaggregation for production is also a context specific matter of modelling judgement. Put differently,
the levels of spatial and sectoral disaggregation are contingent to the problem under scrutiny.
Equilibrium and its perturbation: With these ingredients in place the general equilibrium of the system can be characterised, in which location choices are made and wages, rents, and income levels
determined. The equilibrium may be constrained in the short run because of supply rigidities or factor
immobilities that in the longer run are removed.
4
To understand the workings of these models, it is insightful to worry about how, given some initial
situation (that may either correspond to a short run or a long run equilibrium), the system adjusts to
‘shocks’ to the economic environment. Thinking about the effects of such shocks is also fundamental
from a policy perspective. There are three sorts of shocks to be considered each of which may result in
different kinds of adjustment to the urban or regional system. The first we term ‘location specific’
shocks. That is, some change, which has its direct impact on a single area; for example, the
construction of new houses or the closure of a factory. The second are ‘common’ shocks. These are
shocks, the direct impact of which is felt in many areas. Examples are technological or institutional
changes. The third sort of shock is ‘integrative’. This occurs where the inter-area geography changes,
as when a road is built or telecommunications improved. Such a change will have its own effects,
possibly causing relocation of some activities, as well as changing the economy’s response to other
shocks.
Channels of adjustment: An economic shock changes relative prices and induces changes in quantities – relocations of activity – between areas. There are potentially many channels of
adjustment. Following, say, a positive productivity shock somewhere, firms and workers may relocate,
wages and/or rents may go up, the price of some final and intermediate goods may change, etc. As will
become clear below, the theoretical literature often makes fairly extreme assumptions about the
mobility of goods and workers and the determination of prices. Such simplifications are warranted
when the objective is to clarify some aspect of the workings of the urban/regional system or replicate
some stylised facts. However these simplifications also imply that the channels of adjustment after a
shock are reduced to one or two, whilst most of the others are arbitrarily shut down by the extreme
nature of the assumptions. For policy purposes, however, these channels of adjustment are everything.
For instance, the welfare effect of a negative location specific shock may differ substantially if the
adjustment takes place through a reduction in labour force participation as opposed to out-migration.
In this respect, we do not know of any theoretical model of system of cities or economic geography in
which the participation decision of workers is endogenous. The only channel through which labour
supply is usually assumed to adjust is in and out-migration to and from the area. When trying to
understand why some cities are specialised while others are diversified, this is an acceptable
simplification. When trying to understand how declining industrial cities from Pittsburgh to Liverpool
have adjusted after negative shocks, it is important to consider a variety of channels of adjustment,
including out-migration but also participation decisions, etc.
The main problem when considering a variety of channels of adjustment is that the algebra becomes
very quickly intractable. As we discussed in the introduction, our choice here is to go instead for a
flexible set of assumptions without neglecting the key relationships that govern what happens within
areas and between areas. We first sketch our graphical device before turning to its main applications.
Labour demand: The first key relationship of both urban systems and NEG is the area level aggregate production function relating total output in an area to the local inputs. This relationship has received a
considerable amount of theoretical attention. The main spatial feature that the literature has attempted
to replicate is “agglomeration”, i.e., the concentration of a disproportionate share of economic activity
in a small set of areas. A compelling (and sufficient) condition for agglomeration is the existence of
(aggregate) increasing returns at the area level. Modelling agglomeration in this way is also consistent
with a second major stylised fact: the increase of most measures of productivity per capita with the
size of the local population. Deriving a local aggregate production function, which exhibits increasing
returns without the market structure being degenerate (e.g., a single producer per area), is a significant
challenge.
Despite this, the literature has been fairly successful at proposing microeconomic foundations for such
local increasing returns. These microfoundations typically rely on one of three key mechanisms:
sharing, matching or learning. Sharing mechanisms emphasise how small indivisibilities (like the fixed
costs associated with the production of a new local variety) can be aggregated to generate increasing
returns because a larger local population allows every fixed cost to be spread over a larger number of
customers. Matching mechanisms show how larger markets increase the quality of matches between
economic agents and/or the probability of finding a match. Matching issues are important in a wide
variety of contexts from the local labour market (employer-employee) to the local markets for
intermediate goods. Finally, learning mechanisms aim to show how more frequent direct interactions
between economic agents in denser environments favour the creation, diffusion and accumulation of
knowledge (see Duranton and Puga, 2004, for a review).
A key feature that emerges from the literature on the microeconomic foundations of agglomeration is
that many different mechanisms all lead to a local production function that exhibits increasing returns.
For our purposes, this is a very positive result because one can assume some form of local increasing
returns without having to rely on a specific mechanism. The negative counterpart of this result is that
identifying the precise sources of agglomeration will be difficult (Rosenthal and Strange, 2004). When
concerned with policy, the second caveat is that these local increasing returns derive from specific
market failures. Thus, production is in general inefficient, in the sense that it does not make the best
possible use of local resources. This suggests a role for policy, but the appropriate corrective policies
will depend on the exact mechanism at play: the corrective policies associated with urban knowledge
spill-overs are not the same as those stemming from imperfect matching on the labour market.
very imprecise, it is legitimate to start from such local aggregate production functions and take them
as given.
If we assume that the three primitive factors of production are land, labour and capital and if
furthermore land is perfectly immobile while capital is perfectly mobile, the focus of our attention
needs to be on labour. Rather than on output per worker as function of the size of the local workforce,
it is technically equivalent but more fruitful in terms of interpretation to focus our attention on an
inverse-demand for labour relating the wage of workers to the size of the local labour force. This curve
is represented in figure 1a and we will refer to it as a ‘wage curve’ in what follows. Consistent with
the empirical literature, the wage in area i as a function of the local labour force,w
( )
Ni , is assumed to be increasing in the size of the labour force reflecting the existence of local agglomerationexternalities.
The intensity of the local increasing returns is captured by the slope of the wage curve. Of course, a
neo-classical inverse wage curve would be downward sloping since the amount of land is fixed so the
marginal product of labour decreases with city size and falling land per worker. Whether one should
consider a concave or convex upward sloping curve for w
( )
Ni depends on the specifics of the microeconomic foundations.5 However, this is ultimately an empirical issue. Our concern here is howthe changes in the wage curve affect the equilibrium and how the slope of the curve determines
adjustment. Note that, at this stage, it is easier to assume that workers are either identical or are
horizontally differentiated (i.e., all equally productive but specialised in different activities). Vertical
differentiation (i.e., a workforce where some workers are more productive than others) is ignored for
the time being but its main consequences will be explored later.
Area crowding: The second crucial relationship concerns the various costs associated with having a significant number of households living in the same area and so we will talk of a ‘cost of living curve’.
The main components of the cost of living are the cost of commuting, housing and other consumption
goods. The key question regards the shape of the cost of living curve as a function of the size of the
local population. It seems reasonable to assume that commuting costs increase with population
because a larger population implies longer commutes and more congested roads. Similarly, one
expects increasing population to drive up the cost of land and thus of housing. However, the impact of
area size on the cost of consumption goods is more complicated. A higher cost of land will have
implications for the price of consumption goods throughout the local economy (higher retail costs etc).
5
On the other hand, a larger market offers a wider variety of suppliers without having to import goods
from elsewhere.
Two additional factors, which do not depend on local market size, may also be important in
determining the cost of living. First, external geography plays a role here because importing goods
from other areas may be costly and these costs will be captured by this curve as well. Second, and
particularly crucial from a policy perspective, land-use and planning regulations will also have an
impact since they affect the supply of land.
We represent this curve in figure 1b where we assume that the cost of living increases with population
(the alternative case will be discussed further below).6 For reasons that will become obvious, this
curve is represented with a reversed Y-axis. To be able to use this curve in a simple and tractable way,
we make two simplifying assumptions when drawing it. The first one is that the cost of living is paid
in monetary terms only. The second is that housing consumption per household is fixed. These two
simplifications imply that changes to the wage curve will leave the cost of living curve unchanged. In
reality, commuting costs have both a monetary and a time component so an increase in the wage rate
increases the shadow cost of commuting and thus shifts the cost of living curve downwards on the
figure. By the same token, an increase in the wage rate tends to increase demand for housing and thus
land, making land more expensive. Again, this would shift the cost of living curve downwards. It is
important to note that more formal modelling either ignores these effects or suggests they are second
order and thus do not completely offset the direct effect of a shock to the wage curve. Hence, to keep
the exposition simple, we ignore these effects in what follows. It would, however, be easy (though
cumbersome) to consider them.
Like the wage curve, the position and slope of the cost of living curve reflects a range of market
failures, which means that outcomes may be inefficient. For instance, un-priced urban congestion will
imply an inefficiently high cost of living for any level of population. We will treat these inefficiencies
as given, just like those associated with the microeconomics of production, since the object of this
paper is not their detailed study. This does not, however, prevent the analysis of the broader
implications of, for instance, a reduction in congestion in the centre of a large capital city due to
congestion charging (e.g., London) or the construction of a mass-transit system (e.g., Bogotá). For
such changes, we consider that the main direct effect is to shift the cost of living curve for these cities
upwards. That is, to reduce the cost of living for a given population size. Changes to the supply of land
for housing as, say, local planning regulations are relaxed will shift the curve similarly.
6
The difference between the wage curve and the cost of living curve is represented in figure 1c. This
curve represents the net disposable wage for the area and thus we call it the ‘net wage curve’. In the case of figure 1c, the curve is bell-shaped. This corresponds to the traditional case of urban economics
where agglomeration economies dominate crowding costs for a small population, while the reverse
occurs for a large population. For this to be the case, the slope of the wage curve must be larger than
that of the cost of living curve below a certain threshold, while it is smaller above this threshold. At
this threshold, the net wage reaches its peak (point B in the figure). This peak can be interpreted as
identifying an optimal city size.7
Labour supply: The second curve represented in figure 1c is an inverse labour supply curve. It indicates for any level of real wage, how much labour is supplied in the area. For simplicity, we
temporarily assume that labour supply is a function of the total local population and ignore labour
force participation decisions. In that case, this curve essentially captures the migration response to
local wages. A low level of mobility between areas will be captured by a very steep labour supply
curve (i.e., even large wage differences do not impact much on migration flows), while perfect
mobility will imply a flat labour supply curve (i.e., small wage differences imply large migration
flows). Area specific effects, such as amenities, will shift this curve, with a more attractive area facing
a labour supply curve that is below that of a less attractive area (workers accept a lower real wage but
are compensated by higher amenities).
Equilibrium: The intersection between the labour supply and net wage curves determines the equilibrium of the model. The intersection between these two curves may not be unique. In figure 1c,
the two curves intersect twice (at A and C). The labour supply curve first cuts the net wage curve from
above (at A) and then from below (at C). Point A is not a stable equilibrium. It is easy to see that a
small positive population shock will raise the net wage. In turn, from the supply curve, we see this
higher net wage attracts more workers, which again raises net wages and this process continues until
the area reaches point C. By the same token, a negative shock at point A will lead population and
wages to fall to zero. Turning to the second intersection in C, a similar argument verifies that this
equilibrium is stable. From figure 1c, once we have established the equilibrium population (N*) at
point C, we can trace upwards to figures 1a and 1b to read off the equilibrium wage (w*) and cost of
living (H*), respectively. 7
More accurately, this is the constrained optimal city size. Optimality here is conditional on the local wages being optimal given the population. As shown by Duranton and Puga (2004), unconstrained optimality is unlikely to occur since it requires an efficient production process in the area while, as we discussed earlier in the text, the assumptions needed to generate agglomeration usually require some market failure. Hence
3. A diagrammatic comparison of systems of cities and NEG
In this section, we show that our diagrammatic framework can be used to explain the workings of both
the urban systems and NEG literature. We begin with Henderson’s (1974) system of cities model
before turning to NEG.
Urban systems
Henderson (1974) assumes that there are local increasing returns taking place within industries
(localisation economies). This implies that the wage curve is as in figure 1a but specific to each sector.
Different sectors may show different degrees of increasing returns. Figure 2a represents the wage
curve for two different industries, 1 and 2, as a function of local industry employment. In the figure, industry 2 exhibits stronger increasing returns than industry 1.8 These final goods are assumed to be
freely tradable so that there is no issue of market access.9 We assume that the cost of living depends on
the total workforce in the city and this relationship is drawn in figure 2b.
The first result of the model, that all cities will be fully specialised in equilibrium, follows from the
assumption that increasing returns are industry specific, but the cost of living depends on total
workforce. The reasoning behind this result is the following. Imagine that instead of being fully
specialised, a city has positive employment in both industry 1 and 2. With workers being mobile
between industries, the equilibrium requires wage equalisation between the two industries. The only
possible interior equilibrium must be where the two curves intersect. Assume then a small shock,
which moves employment to industry 1 at the expense of industry 2. Such a shock will imply a higher
wage in industry 1 and a lower wage in industry 2. This in turn will imply more workers moving from
industry 2 to industry 1, etc. The only stable equilibrium must thus involve the full specialisation of
each city in either industry 1 or industry 2.
As discussed above, provided that at some point the marginal increase in cost of living begins to
dominate the marginal increase in wages in industries 1 and 2, the net wage curve will be bell-shaped
for both industries. These two curves are represented in figure 2c. Note that optimal city size is larger
for industry 2, with the stronger increasing returns. Assume that labour is perfectly mobile across
8
Note that in figure 2a, the wage curves for the two industries cross. Assuming that both goods are indispensable then, in equilibrium, prices of the two goods must adjust to ensure that this is the case. If the two curves did not cross, one activity would always pay more and thus attract all the workers and the other good would not be produced.
9
cities so that the labour supply curve is flat. This implies that the short-run equilibrium will be given
by points A and B, with net wage the same in both cities and the population NA and NB for the cities specialised in industry 1 and 2, respectively.
At this stage, two important properties must be noted. First, all cities will be too large with respect to
their constrained optimal city size following the same stability argument as previously.10,11 Second, all
cities with the same specialisation must be of the same size. It is interesting to note that this prediction
receives mild empirical support (Henderson, 1988).
The crucial issue with regard to long run equilibrium concerns the mechanism for city creation. In the
long run, the number of cities is not fixed but instead can vary depending on the action of land
developers. The model assumes that these developers can create cities of a size of their choosing and
can use fiscal instruments to appropriate rent to finance their activity. The optimal strategy of a
competitive land developer is thus to create a city of optimal size and tax the workers the difference
between the real wage that the city offers and the real wage that can be obtained elsewhere. As
workers move to the newly created cities, they leave old established cities. A decline in the size of the
latter will increase the net wage they can offer to their workers. In turn, this dampens the incentive for
further new cities to be developed. In equilibrium, the free-entry of competitive developers will imply
that all cities will reach their optimal size.
This is not the end of the story, however. Note that in figure 2c the peak of the industry 2 curve (the
maximum real wage payable in an optimally sized industry 2 city) lies above the peak of the industry 1
curve. That is, cities in industry 2 can offer a higher wage at their constrained optimal size. However,
these differences cannot persist in equilibrium with competitive land developers because they imply
higher returns for building industry 2 cities. These higher returns induce the development of relatively
more new cities specialised in industry 2. Thus output in industry 2 increases more than output in
industry 1 and this causes the relative price of industry 2’s final good to decline. This decline in price,
pushes down the demand for labour in industry 2 and so the wage curve moves downwards. This
10
It is important to note that cities are oversized only with respect to their constrained optimal size (i.e., the “optimal” size, which ignores inefficiencies in both production and housing/commuting). If for instance production is inefficient in figure 2a, it may well be the case that the first-best wage curve is above the equilibrium wage curve with a steeper slope. This steeper first-best wage curve would imply an unconstrained optimal city size larger (and possibly much larger) than the constrained optimal size. Then the comparison between the fully optimal city size and the equilibrium city size is ambiguous since there are two distortions pushing in opposite directions. On the one hand, without a market for cities, equilibrium city size is too large. On the other hand, all the benefits from agglomeration are unlikely to be exhausted, which implies equilibrium city size being too small. As discussed in the text, in reality, we also expect important inefficiencies regarding the determination of the cost of living. Such inefficiencies are likely to increase the gap between the constrained and unconstrained optimal city size.
11
process continues until optimal city size is the same for both types of cities. This long run equilibrium
is shown in figure 2d. Developers make zero profit for both types of cities, all types of cities reach
their optimal size and workers receive the same net wage everywhere. Finally, cities export the good
they produce and import all other goods.
This model is still a landmark in the literature, providing the first consistent model of an urban system
in which cities arise endogenously from a tension between agglomeration economies and urban
crowding while interacting together through trade. The usefulness of this model for policy purpose
remains, however, limited. Consider for instance a negative productivity shock in a given city (which
reduces the marginal productivity of labour by a given proportion). The net wage at the optimal city
size will end up below that offered in other cities. Workers will move out of this city and a land
developer will choose an empty site to develop another city of optimal size. Alternatively, a negative
common productivity shock affecting all cities of a given type will lead to cities of smaller size and,
depending on the substitutability between goods, more or less cities of that type. These examples
clearly show that the ultimate margin of adjustment in the model is the number of cities. In turn,
adjusting the number of cities depends crucially on the assumption of perfect labour mobility. Clearly,
cities do not enter and exit the urban landscape like this, particularly in European countries where the
activity of land developers is severely restricted.12
Despite its drawbacks for policy analysis, this model has served as a starting point for much of the
subsequent theoretical work on urban systems (see Duranton and Puga, 2000, for a survey of this
literature). Unfortunately, most of the extensions of this framework are concerned with only two
issues: providing more detailed or alternative microeconomic foundations for the wage curve and
generating more realistic predictions for the composition of economic activity in cities, since full
specialisation is obviously too extreme. These extensions, however, do not tackle the problem of the
entry and exit of cities (Helsley and Strange, 1997, and more recently, Henderson and Venables, 2004,
are rare exceptions) or the issue of imperfect labour mobility. Later in the paper, we consider the issue
of labour mobility as well as a number of policy relevant extensions, but for now we leave the urban
systems literature and turn our attention to NEG.
12
New Economic Geography
Despite very significant modelling differences, the core models of NEG can be explained with the
same graphical device. This is because, as we argue below, a wage curve and price index (i.e., cost of
living curve) lie at the heart of NEG models. As above, the difference between the nominal wage and
the cost of living curve then defines a net wage curve and assumptions on labour mobility between
areas allow one to derive the long-run equilibrium.13 The use of our graphical device is, however,
made more difficult in NEG because in some core models (e.g., Krugman, 1991), the wage equation
and the price index do not admit closed-form solutions. To get round this difficulty, we can either rely
on numerical solutions for specific parameter values or instead use variants of these core models that
can be solved analytically (Ottaviano, Tabuchi and Thisse, 2002, and Forslid and Ottaviano, 2003).
Models in the line of Krugman (1991) consider two areas and two sectors.14 For ease of exposition in
what follows, we appeal to NEG’s roots in international trade and refer to the first area as ‘Home’ and
the second as ‘Foreign’. The first sector produces some homogenous good under constant returns. To
simplify the derivation of the model, this good is assumed to be perfectly tradable (so that its price is
equal in both areas and can be normalised). This sector is often referred to as ‘agriculture’ or identified
with some traditional good for which workers are immobile both geographically and sectorally (i.e.,
they always work in this homogenous good sector). Given these mobility assumptions, the wage in
this sector can then be derived directly from the normalised price of the good it produces. For
empirical and policy purpose, it is enough to assume the existence of some autonomous or residual
demand in each area (e.g., civil servants, pensioners, etc) so that this sector can remain in the
background, while attention focuses on ‘manufacturing’. Manufacturing consists of a number of firms,
each operating with internal economies of scale and producing their individual variety of differentiated
product. This sector is monopolistically competitive as in Dixit and Stiglitz (1977). Assumptions on
the substitutability between goods ensure that all consumers demand all varieties and so each firm
sells its output in both areas. Thus if manufacturing is operating in both areas, there is ‘intra-industry
trade’. However, the presence of transport costs means that firms’ sales have a ‘home-market bias’.15
Manufacturing workers are mobile and go to the area that offers them the highest utility.
13
Note that for the ease of presentation the price index is subtracted from the wage, instead of dividing the wage by this index. For this to be technically correct, both the wage curve and price index should be drawn on a log scale.
14
The properties below are those of Forslid and Ottaviano’s (2003) model, which differ only minimally from Krugman (1991).
15
The wage of the workers in the manufacturing sector in an area is represented by the wage curve in
figure 3a. The shape of the wage curve is the result of a subtle trade-off between two opposing forces:
a crowding effect on the product market and a home-market effect. The resolution of this trade-off and
hence the slope of the curve, depends on the level of transport costs. For high transport costs, the wage
curve is downward sloping (the plain curve), while for low transport costs it is upward sloping (the
dashed curve).
To better understand this, consider Home (symmetric arguments hold for Foreign) and assume that
employment in the manufacturing sector is arbitrarily small in Home (and thus most manufacturing
activity takes place in Foreign). Note that with high transport costs, manufacturing goods produced in
Foreign will be expensive in Home. Hence, not only does Home have relatively few producers but
these producers are also protected from imports by high transport costs. They can thus charge high
prices, which imply high wages for the workers employed in manufacturing in Home. Hence, when
the manufacturing workforce in Home is small and transport costs are high, Home pays high
manufacturing wages. Now, continue to make the same assumption on the shares of employment in
the two different areas, but assume that transport costs are very low. With low transport costs, the
protection enjoyed by manufacturing in Home is eroded. In turn, this leads to lower manufacturing
prices and thus lower wages in Home. Hence, when the manufacturing labour force is small in an area,
local wages are increasing in transport costs.
Consider now the effect on the wage of an increase in the size of the manufacturing workforce in
Home. There are two offsetting effects. First, a larger workforce in manufacturing means a larger
income in Home. Given that transport costs are positive, this extra demand will disproportionately
benefit Home producers. This is the home market effect and it tends to increase wage. Note that the
home market effect is weaker when transport costs are low. This is because lower transport costs make
it easier to import foreign goods. Consequently, when Home income increases following an expansion
of the manufacturing workforce, a greater share of that increase will be devoted to manufacturing
goods produced in Foreign when transport costs are lower.
The second effect of a larger manufacturing work force is that there are more firms, which crowds the
product market. When transport costs are high, the demand from Foreign will be low and the market
faced by the Home firms will essentially boil down to the Home market. Then, a larger manufacturing
workforce implies more Home firms but more Home firms simply lead to a more crowded local
product market and this lowers prices and wages. This is the market crowding effect mentioned above.
Note that the market crowding effect is also attenuated for lower transport costs. This is because lower
transport costs make it easier to export manufacturing goods so that an expansion of the manufacturing
transport costs, crowding dominates and the wage curve is downward sloping (the solid line in figure
3a). In contrast, when transport costs are low, the home market effect can dominate leading to an
upward sloping wage curve (the dashed line in figure 3a).
The reasons for this reversal are quite subtle. We have already shown above that when the
manufacturing labour force is small in an area, local wages are increasing in transport costs. We now
show that when the manufacturing labour force is large in an area, this relationship is reversed
implying that the wage curve slopes up for low transport costs and down for high transport costs. To
see this, it is easiest to take a specific example. Thus, assume that Home employs 50 workers in
agriculture and 100 workers in manufacturing, while foreign employs 50 workers in each sector.
Consider first the market crowding effect, which depends on the number of manufacturing firms in
Home relative to the size of the market for those firms. As transport costs decline, the manufacturing
sector in Home gradually gains a better access to 100 consumers. Hence, as transport costs decline
from infinity to zero, the market for manufacturing firms located in Home increases from 150 to 250, i.e., by 66%. Consider now the home market effect, which depends on the share of home
manufacturing firms in the purchases of Home consumers. As transport costs decrease, workers in
Home spend a higher fraction of their manufacturing expenditure on imported goods. More
specifically, when transport costs are infinite, Home consumers buy their manufacturing varieties from
the home manufacturing sectors, that is from 100 manufacturing workers. When transport costs are
zero, they can buy their manufacturing goods from manufacturing in both areas, that is from 150
manufacturing workers. Hence, as transport costs decrease from infinity to zero, the manufacturing market for consumers in Home increases by 50%. Put differently, as transport costs decline, the market for Home producers increases faster than the market for home consumers. Thus market
crowding decreases faster than the home market effect. For sufficiently low transport costs the home
market effect dominates the crowding effect and Home will pay higher wages. That is, for sufficiently
low transport costs, local wages in large areas are decreasing in transport costs. Remembering that the
relationship between local wages and transport costs is reversed for small areas, we see that for low
transport costs the wage curve must be upward sloping while it must be downward sloping for high
transport costs.
What about the cost of living? By assumption, NEG neglects issues relating to land and housing and
so the cost of living in an area will be determined only by the price index for the imperfectly traded
(i.e. manufacturing) goods. When the manufacturing workforce in an area is small, the share of
imported goods is large, which puts upward pressure on prices as these goods are subject to transport
costs. As local market size increases, local prices are pushed downwards. Consequently, the cost of
Lower transport costs have two implications for the cost of living curve. First, it is obvious that,
conditional on the size of the area, the price index is lower when transport costs are lower. Second,
and slightly less obvious, the lower transport costs the less important is own area size in determining
the price index and so the flatter is the cost of living curve. Thus in figure 3b the dashed line gives the
cost of living with low transport costs, the solid line with high transport costs.
In figure 3c, we report the resulting net wage curve, i.e., the difference between the wage and the cost
of living, for both high and low transport costs. In each case, there is a symmetric curve for the other
area since the sum of the population of both areas is a constant. Thus we can draw the curves for both
areas on the same figure (with a thick line for Home and a thin line for Foreign), the distance between
the two vertical axes being equal to the total population. For high transport costs, the relevant net wage
curves are represented by the solid lines which are decreasing: net wage declines with the size of the
local manufacturing population since both the nominal wage and the cost of living increase. Both
curves intersect for areas of equal size at A, which represent the symmetric equilibrium. The
equilibrium is stable in this case. To see this, imagine shifting a worker to Home from Foreign. The
net wage falls in Home and rises in Foreign so that the worker wishes to move back to Foreign and so
symmetry is restored. In contrast, for low transport costs, the relevant net wage curves are represented
by the dashed lines, which are now increasing. The symmetric equilibrium at B is now unstable.
Again, to see this, imagine moving one worker to Home from Foreign. In contrast to the situation with
high transport cost, this perturbation raises the net wage in Home, thus attracting more workers and
further raising the net wage. This suggests that with low transport costs, the only stable equilibria
involve concentration in either Home or Foreign. Of course, given that the two areas start out
symmetric, we cannot say which of these two equilibria will be reached.
One of the strengths of this class of model is that it allows investigation of integrative shocks. Indeed,
in his original paper Krugman (1991) focuses on the question of what happens to the equilibrium as
transport costs are reduced. The answer is given by the comparison between the two kinds of
equilibria in figure 3c. If transport costs are high, then manufacturing is equally divided between the
two areas, essentially because of the need to supply the immobile consumers. As transport costs fall, it
becomes easier to supply all consumers from a single area. At low transport costs the symmetric
equilibrium is unstable, and the stable equilibria require all manufacturing activity be concentrated in
one of the two areas (with the equilibrium being at either point C or C’, depending on whether the
agglomeration occurs in Home or Foreign, respectively).16
16
The model demonstrates in a particularly sharp way how concentration can arise from market access
effects, even without any externalities within the manufacturing sector. The work of Krugman (1991)
has served as point of departure for a large literature (see Ottaviano and Thisse, 2004, for a survey).
Before turning to these extensions, a comparison between the two frameworks analysed so far is
warranted.
Comparing the two canonical models
As we discussed in the introduction, the approaches based on urban systems and NEG are often
perceived as disjoint and commonly studied and further developed separately, often by different
people. Although the approaches do differ, the previous two subsections make it clear that the
differences between urban systems and NEG are about the key assumptions of the models rather than
the nature of the models themselves. In a nutshell, the core model of urban systems relies on local
agglomeration effects and local congestion effects in the context of a large number of areas. NEG
instead relies on firm level increasing returns and the existence of transport costs between a small,
discrete number of areas.
These assumptions translate into differences for the shape of the three mains curves: the wage curve,
the cost of living curve, and the labour supply curve. The wage curve is unequivocally increasing in
the urban systems approach. Because of all sorts of local externalities (technological spillovers, thick
labour market effects and input-output linkages) it is assumed that increasing returns occur in each
area. In contrast in NEG the wage curve may be increasing or decreasing. This is because, as we have
shown, the subtle trade-off between a crowding effect (a larger workforce leads to an increasingly
crowded local market) and a home market effect (a larger workforce implies a larger local market
which benefits all local firms) depends on the level of transport costs. Even when the wage curve is
upward sloping, there are no pure local externalities in the NEG model, only pecuniary effects that
occur because of transport costs and increasing returns to scale at the firm level.
Ultimately, the most appropriate shape for the wage curve is an empirical question. However, it is
important to note that a priori, the assumption made by the urban systems literature would appear to be
relevant at smaller spatial scales where local externalities are expected to play a role. By contrast, the
NEG assumptions are better suited to larger spatial units (regions, countries or even groups of
countries) for which long distance market interactions are expected to play an important role while
short distance effects become of secondary importance.
Turning to the cost of living curve, there is again a sharp contrast. The cost of living increases with
wage curve, the shape of the cost of living curve is an empirical issue. However, the spatial scale at
which these theories are applied is likely to matter for this curve as well. With a more urban focus, the
literature on systems of cities pays a lot of attention to rising land and commuting costs. Instead, NEG,
with its more regional perspective, views commuting and housing costs as second order issues
compared to the importance of market access and its impact on the price of consumption goods.
The final curve to consider is the labour supply curve. Both sets of models assume that manufacturing
workers are perfectly mobile between areas and that they will move until net wages are equalised
across areas. There is one subtle difference in that assumptions on the spatial distribution of immobile
workers fixes a-priori the number of spatial units in NEG approaches while the number of areas is
determined endogenously in the urban systems approach. However, the role of the number of spatial
units turns out to be relatively unimportant. As shown by Fujita et al. (1999), the results of the core
NEG model generalise to a larger number of regions.17 From our diagrammatic exposition above, it is
clear that the urban systems approach can readily be generalised to a small, discrete number of cities
(see Papageorgiou and Pines, 1999, for more on this).
Pulling all of this together, we would argue that there is no inherent contradiction between the urban
system approach and NEG: the latter is trying to explain broad trends at large spatial scales while the
former attempts to explain “spikes” of economic activity. Clearly, bringing these two approaches
together in a unified framework is an important goal for future research.18 This is important because
distinguishing empirically between a “trend” and a “spike” is not easy, especially as a series of spikes
can imply a trend when none may be present.19
4. Using our framework with existing extensions of the system of cities and NEG approaches
In this section, we use our framework to consider a number of extensions to the urban systems and
NEG models that have been proposed in the literature. We start with two extensions to the NEG
approach, before turning to extensions to the urban systems model.
The rise and fall of regional inequalities: adding housing
The core NEG framework can be enriched by either adding more features on the production side
(changing the wage curve) or more features on the dispersion side (changing the cost of living curve).
17
However, for policy purpose the number of areas matters for the number of winners and losers and the magnitudes of the gains and losses as the spatial economy adjusts to shocks.
18
See Fujita, Krugman, and Mori (1999) and Tabuchi (1998) for two very different early attempts.
19
Regarding the cost of living curve, one particularly natural extension is to impose some constraints on
land supply by assuming that housing is imperfectly elastically supplied so that house prices are
increasing with population.20 This extension can easily be incorporated in to our diagrammatic
framework by recognizing that the cost of living curve will now depend on both the cost of
consumption goods and the cost of housing. As the size of the area rises, the price of consumption
goods falls (as discussed above) but this fall is offset by rising house prices. At some point, we would
expect rising house prices to dominate and this implies a bell-shaped cost of living curve. Perhaps the
clearest way to picture this is to add together the cost of living curve in figure 1, which captures house
prices and congestion, with the curve from figure 3, which captures the price index effect. This
non-monotonic cost of living curve is represented in Figure 4b. As with the core NEG model, falling
transport costs will have two effects on this cost of living curve. First, lower transport costs shift the
curve up. Second, as transport costs decline, the first part of the bell shape for the cost of living
becomes flatter as market access matters less and less. At very low transport costs, market access is
essentially irrelevant and the cost of living curve is monotonically increasing in the size of the
manufacturing work force.
For high transport costs, combining the wage curve from the core NEG model with a bell-shaped cost
of living curve leaves the net wage curve essentially unchanged compared to figure 3, so symmetric
areas is still the only stable equilibrium. It is represented by point A in figure 4c. For intermediate
transport costs, a bell-shaped cost of living curve implies a bell-shaped net wage curve consistent with
some, but not complete agglomeration of manufacturing workers in a single area. Note that this
extension leads to a net wage curve that is similar to that in the urban system approach. This is
particularly interesting given that the forces driving these two curves are very different. In this case,
the symmetric situation is no longer a stable equilibrium. Instead the two stable equilibria at B and B’
involve some agglomeration of manufacturing in one area or the other. Finally, when transport cost are
very low, the net wage curve now intersects the net wage curve for the other area in its downward
sloping region. The symmetric situation in C is again a stable equilibrium.21
Once again, we can work through the thought experiment of allowing transport costs to fall and think
what this implies for equilibrium. When transport costs are very high, manufacturing workers are
evenly distributed across areas because it would be too expensive to serve the agricultural demand of a
‘peripheral’ area (no manufacturing) from an agglomerated ‘core’. Then, for intermediate transport
20
Helpman (1998) present a formal model incorporating this assumption but changing other key assumptions of the core Krugman model. In the text, we prefer to use our graphical framework to assess what happens if housing costs are added to the core Krugman model.
21
costs, it pays for manufacturing workers to agglomerate and economise on transport costs while
serving peripheral farmers from the core. Finally, for low transport costs, high housing costs in the
core become dominant. The effects of market integration on the location of activity are therefore
ambiguous in this type of framework. At relatively high levels of transport costs a reduction in barriers
promotes spatial disparities. At lower levels it permits the dispersion of activity and a narrowing of
disparities. This is an example of the famous bell-shaped effect of market integration on regional
inequality.
The rise and fall of regional inequalities: adding input-output linkages
A similar type of outcome can be obtained from a very different extension of NEG. Following Puga
(1999) and Krugman and Venables (1995), assume workers are mobile across sectors (agriculture and
manufacturing) but not across areas. This last assumption allows us to simplify our diagram since
labour market equilibrium only requires that agriculture and manufacturing workers get paid the same
nominal wage rather than the same net wage, since all types of workers in an area will face the same
cost of living.22 The agricultural good is freely tradable across areas and is produced using labour and
land. The wage in agriculture as a function of manufacturing employment is represented by the dashed
upward-sloping curve in figures 5 a-c. This line slopes upwards because increased manufacturing
employment in an area implies lower agricultural employment, higher land to labour ratios in
agriculture and thus higher agricultural wages.
The manufacturing sector is as in Krugman (1991) with the added complication that existing varieties
are also used as intermediate inputs by manufacturing producers. This implies that the shape of the
wage curve for manufacturing will be the outcome of a trade-off between three different forces. First,
and as in Krugman (1991), market crowding still occurs: as the size of the manufacturing sector
increases locally, the price index declines. Under free-entry, a decline in the price index implies that
the maximum wage that producers can pay also declines. Second, there is also a home-market effect as
in Krugman (1991). Since workers are immobile geographically, the workings of this effect are
slightly different from that in Krugman (1991). A large manufacturing sector locally implies a larger
market for manufacturing goods because the manufacturing sector consumes its own varieties as input,
whereas in Krugman (1991) the effect was driven by the increase in the number of workers. Third,
there is an additional local increasing return working through cost linkages: A larger local
manufacturing sector means a greater number of varieties and a lower price index for manufactured
goods, which are used as inputs in to the manufacturing process as well as for final consumption.
Coupled with free entry, this works to partially offset the effect of market crowding on wages. Just
22
like in Krugman (1991), the crowding effect dominates when transport costs are high while the home
market and cost linkage effects become more important for lower transport costs.
The plain curve in figure 5a represents the wage curve for manufacturing in the case of high transport
costs. Consider an increase in the manufacturing workforce starting from zero employment in
manufacturing. Cost linkages and the home market effect first dominate before crowding kicks in and
lowers manufacturing wages as the manufacturing workforce expands. The equilibrium occurs at point
A. The plain curve in figure 5b represents the manufacturing wage curve for intermediate transport
costs. In this case, the home market effect reinforced by cost linkages dominates the crowding effect
as manufacturing expands. There are now three equilibria at points B1, B2, and B3. The equilibrium in
B2 is unstable so that area 1 is either at B1 while area 2 is at B3 or area 1 is at B3 while area 2 is at
B1. Compared to A1, B3 implies a larger manufacturing workforce.23 Finally the plain curve in figure
5c represents the manufacturing wage curve for low transport costs. In this case, cost linkages are
much weaker so that the manufacturing wage curve is much flatter than in the previous case. There is
a unique equilibrium at C. This equilibrium implies a more even distribution of manufacturing than for
intermediate transport costs. The reason behind this is that concentration of manufacturing in one area
bids up the wage in that area while it depresses the wage in the other area. When transport costs are
low, manufacturing in the core area has to compete with manufacturing in the peripheral area. Since
workers are the same, the only source of wage difference arises from the costs of importing
intermediates and exporting output. As transport costs decline, the wage difference between the two
areas has to decline, which implies manufacturing moves back to the periphery. The bell-shaped effect
of market integration is again obtained.
Sorting
As shown by Combes, Duranton, and Gobillon (2005), workers tend to sort across cities according to
observed and unobserved characteristics. More generally, skilled workers are over-represented in large
cities while unskilled workers tend to live in smaller cities (Peri, 2002). Inspired by the analysis of
Abdel-Rahman and Wang (1997), it is possible to show how such a pattern can emerge using our
diagrammatic framework. The starting point of the analysis is to assume that workers with higher
skills benefit more from agglomeration effects. If, for instance, agglomeration has a multiplicative
effect on individual productivity (as routinely assumed by empirical studies), the benefits from
agglomeration will increase with skills. Graphically this implies two wage curves, one for skilled
workers and another for unskilled workers. This is represented in figure 6a. After taking into account
the cost of living curve, the net wage curves for these two groups of workers are represented in figure
23
6c.24 Unsurprisingly the net wage curve for skilled workers is above that for unskilled workers. With
both groups of workers being perfectly mobile and assuming that skilled workers enjoy a higher
reservation level, the two labour supply curves are also represented in figure 6c.
The equilibrium is represented by points A and B for skilled and unskilled workers. It can be verified
that at this equilibrium, skilled workers have no incentive to move to a smaller unskilled city while
unskilled workers would enjoy a lower utility by moving to a larger skilled city. Thus, sorting by skill
is an equilibrium outcome.
Changes in specialisation and the structural transformation of cities
The extension of the urban system framework proposed by Duranton and Puga (2005) is useful for
thinking about the transformation of the very largest city economies during the de-industrialisation
period. This model was motivated initially by changes in the US urban system which saw cities
shifting from being predominantly specialised by sector to being specialised by function (e.g.,
production, management, etc). To model this, Duranton and Puga (2005) use the Henderson (1974)
framework but allow the spatial organisation of firms to be endogenous.
Firms, in each sector, require both sector-specific inputs (e.g., car parts for car producers) as well as
business services for their headquarters. There are agglomeration economies in all sectors including
business services. Firms face a trade-off between spatial integration of both headquarters and
production facilities and the spatial separation of these two functions. If firms decide to locate their
production facilities and headquarters separately, then both parts of the operation can fully benefit
from the relevant agglomeration externalities (sector specific inputs for the production facility,
business services for the headquarters). In contrast, spatial integration allows firms to manage the
interaction between production facilities and headquarters more efficiently because they save on
communication costs between the two units, but this comes at the cost of more expensive inputs to
both parts of the operation (because of the negative reciprocal crowding that both activities generate in
the city).
When communication costs are high, it is very costly for firms to spatially separate their headquarters
from their production plant. Consequently the demand for labour of spatially disintegrated firms is
low. As a result, a functionally specialised city will pay low wages, captured by the bottom wage
curve in figure 7a. In particular, this wage curve will be below that of a city specialised by sector
24
(medium wage curve in figure 7a). However, when communication costs are low, it is beneficial for
firms to spatially separate their facilities and, as a result, cities will be specialised by functions,
production or business services. The advantage of functional specialisation is that headquarters
(regardless of their sectors) will be able to exploit more fully the economies of scale in cities
specialised in business services and management activities. For low communication costs, the wage
curve of functionally specialised cities is represented by the top curve in figure 7a.
Using these wage curves, together with the cost of living curve represented in figure 7b, we draw, in
figure 7c, the net wage curve for all three types of cities. It is clear that functionally specialised cities
are not viable for high communication costs (the net wage curve does not intersect with the labour
supply curve). For low communication costs, functionally specialised cities are viable and the
equilibrium size of a city functionally specialised in management and business services when communication costs are low is larger than that of a city specialised by sector.
If we accept the idea that technological progress in the last 40 years has made it easier to manage
production activities remotely (cheaper telecommunication, intranets, etc), we can shed some light on
the structural transformation of the economy of cities such as London, New York or Paris. In the case
of London, for instance, the structural transformation that took place between the 1960s and the 1990s
can be understood as a successful transition from an integrated city with a production structure
encompassing a range of manufacturing sectors, management, government, and the handling of goods
to a functionally specialised city concentrated in business services, management, and high-end
government activities. In figure 7c, this corresponds to a shift from A to B. This increased
concentration of high value-added activities with large agglomeration economies in London may have
taken place to some extent at the expenses of many cities from the rest of the country. As shown by
Duranton and Puga (2005), if we assume (in line with empirical evidence) that business services
benefit from stronger localisation economies than other industries, business centres will benefit from
stronger agglomeration economies than cities combining production and management and in turn the
latter will enjoy stronger economies of scale than purely production cities. Referring back to figure 2a,
and holding communication costs constant, one can think of business services as industry 2,
production activities as industry 1 with integrated cities as a hypothetical industry 3 with the net wage
curve lying between industry 1 and 2. This implies that cities loosing their management functions also
have a lower optimal size. In the London case, this implies that the counterpart of the growth of
London was the population decline of cities in Northern England.
In a world of perfect mobility, this structural transformation would only imply higher wages for all
workers because a more efficient urban organisation increases productive efficiency. In turn, with
